AdarshSingh7647/TabRankSingleTableCoTCond
AdarshSingh7647/TabRankSingleTableCoTCond is an 8 billion parameter generative listwise reranker built on Qwen3, designed for table retrieval tasks. This model, trained on the NQ Tables dataset, efficiently orders candidate tables from most to least useful for answering a given question. It processes all candidates in a single prompt and generates the full ranking as a JSON object, making it particularly effective for single-table reasoning conditioned on short reasoning hints.
Loading preview...
TabRankSingleTableCoTCond: A Generative Listwise Reranker for Table Retrieval
AdarshSingh7647/TabRankSingleTableCoTCond is an 8 billion parameter model based on Qwen3, specifically engineered for listwise reranking in table retrieval. Its core function is to take a question and a set of candidate tables, then output a ranked list of tables from most to least relevant for answering the question.
Key Capabilities & Features
- Single-Call Reranking: Processes all candidate tables within a single prompt and generates the complete ranking in one generation, outputting a JSON object with ranked table positions.
- Reasoning Conditioned: This specific variant is trained to produce rankings conditioned on a short reasoning hint provided in the prompt, enhancing its accuracy and effectiveness.
- Optimized for Single Table Retrieval: Trained exclusively on the NQ Tables training split, focusing on questions requiring single-table answers.
- Performance: Achieves strong evaluation metrics across various table question answering benchmarks, including SQA, TAT QA, HybridQA, and TabFact, with mean recall@10 of 0.812 and mean MRR of 0.755.
When to Use This Model
This model is ideal for applications requiring efficient and accurate reranking of candidate tables for question answering, especially when dealing with single-table retrieval scenarios. Its ability to incorporate reasoning hints makes it suitable for tasks where contextual guidance can improve ranking quality. Developers can integrate it using vLLM or Transformers for table relevance tasks.